On 02. 11. 26 11:13, Libor Peltan wrote:
Hi Geoff,

On 01. 11. 26 20:09, Geoff Huston wrote:

On 31 Oct 2026, at 09:51, Joe Abley <[email protected]> wrote:

My opinion is that the DNS needs less UDP, not more of it. I am not in favour of band-aids that prolong the tolerance of transport protocols that constrain response sizes.

I think we should be doing everything we can to move the mainstream to (ideally encrypted) stream-based transport protocols which have no response size constraints beyond those of the DNS message format, transport protocols that have no need for protocol modifications of the kind you describe above.

The future of UDP transport should be best-effort support for legacy devices.

Over in the world of paying for infrastructure "cheap" rules. I have listened to a number of presentations over the years that note that a DNS server's throughput drops by some 60% when the same query load is presented over TCP rather than UDP. My experience with DNS servers supports this quite radical drop in capacity (in my case throughput dropped by 90%!). So where does the money come from to upgrade DNS infrastructure to support a transition to TCP everywhere in the DNS?

Yes but also among DNS-over-UDP traffic, only a fraction of the volume is _legitimate_ DNS queries. The rest is noise and attacks.

_Because_ of UDP, the infrastructure also has to be expensively overprovisioned, in order to sustain those attacks, together with mitigating amplification attacks on 3rd-parties.

Regarding TCP, I think that the existing DNS infrastructure is already scaled enough to manage DNS-over-TCP for all legitimate traffic.

The remaining question is, if it is also ready for TCP-specific attacks. And what is the price.


Maybe things would be different if we had started with DNS over TCP at the outset, but now we are in the UDP room there is a price to be paid to move to TCP and no clear view as to who and how we pay that price.

I totally agree with Joe and Libor (a rare event!)

Speaking of DNS over-provisioning, have a look at
https://stats.nic.cz/dashboard/en/Traffic.html
so we can discuss some data. Scroll to chart at the bottom.

Average QPS in 'steady state' is below 50 k/sec. Even _my laptop_ can handle whole CZ TLD on a _single CPU core_, even with venerable BIND! And still there will be capacity to spare!

All the over-provisioning (which, I believe, is over 400 GBps for a single CZ site...) is for attacks over UDP, which cannot be blocked easily because they lack source validation.

That's _ridiculous_ amount of money spent on running a ridiculously insecure protocol.

Let's finally get over it and fix the protocol / move to better one instead of adding layers of hacks upon layers of existing hacks. If we do that there will be plenty _excess_ money which does not need to be burned on over-provisioning anymore.

--
Petr Špaček
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to